New Evidence on Labor Market Flows and the Hiring Process

Transcription

1 New Evidence on Labor Market Flows and the Hiring Process Steven J. Davis University of Chicago and NBER September 2011 HEC Montréal

2 Two Papers p Labor Market Flows in the Cross Section and Over Time, August 2011, NBER Working Paper No , Journal of Monetary Economics, forthcoming January p The Establishment-Level Behavior of Vacancies and Hiring, NBER Working Paper No , revision coming soon. Both with Jason Faberman and John Haltiwanger.

3 Overview of Part I 1. Examine joint behavior of worker flows and job flows in the cross section (CS) of employer growth rates. 2. Interpret joint behavior in light of search and matching theories. 3. Use statistical models of worker flows in the CS to explain aggregate flows. How much gain? 4. Combine statistical models with administrative data on distribution of establishment growth rates to construct synthetic measures of hires, separations, quits and layoffs

4 Two U.S. Data Sets Job Openings and Labor Turnover Survey (JOLTS) Monthly sample of 16,000 establishments covering nonfarm economy. Rotating panel design. Each establishment reports employment, hires, quits, layoffs, other separations, and (end-of-month) vacancies Our micro sample covers Jan-2001 to June-2010 and includes all establishments with data for all three months in a quarter. Business Employment Dynamics Data (BED) Quarterly administrative data on nearly all US establishments in the private sector Micro Data cover 1990Q1 to 2010Q2 Micro data are longitudinally linked allows calculation of establishment-level growth (i.e., job flows)

5 Quarterly Worker Flows in the Cross Section, United States, Pooled JOLTS Sample, Percent of Employment Hires Layoffs Total Separations Quits Employer growth rate measured as g t = E t! E t!1 (1/ 2)(E t + E t!1 ) "[!2, 2] Establishment- Level Employment Growth (Pct. of Employment)

6 Theory Sketch p Search models in the spirit of Mortensen and Pissarides (1994) but with multi-worker firms n E.g., Cooper-Haltiwanger-Willis (2007), Elsby-Michaels (2011) n Iron link of hires to job creation & separations to destruction p Learning about match quality as in Jovanovic (1979, 1985) and Moscarini (2005) n Pries & Rogerson (2005) is a hybrid of MP and learning p On-the-job search with match-specific productivity and aggregate fluctuations (Barlevy, 2002) n Workers are more likely to quit bad matches when aggregate conditions are strong p Employer search with persistent idiosyncratic firm profitability (Faberman & Nagypal, 2009) n Workers are more likely to quit employers with low productivity and slow growth (an abandon-ship effect)

7 Standard Model with Iron Link Implications Consider an MP model with multi-worker firms Cooper-Haltiwanger-Willis (2007, CHW) Hires, vacancies and layoffs are endogenously determined subject to fixed and variable costs of posting vacancies and layoffs Firms face aggregate and idiosyncratic profit shocks Quit rate is exogenous and uniform Workers are ex ante homogenous Frictional search as in other MP models Write employer-level growth (hires separations) as e it e i, t 1 = h = η l qe it it i, t 1 ( Ut, Vt ) vit lit qei, t 1 where ( ) is the job-filling rate, which depends on aggregate unemployment (U t ) and vacancies (V t )

8 CHW Model Properties Inaction at q (mass point) p Movements in aggregate hires and layoffs arise entirely from shifts over time in CS distribution of employer growth rates. p Adjustment costs and shock properties affect the shape and location of growth rate distribution, but not the iron link.

9 Relaxing the Iron Link Simplest extension of CHW model: Quit rate remains exogenous but varies procyclically q t = q(g t ) where G t = aggregate employment growth Iron link continues to hold in a given cross section, but time variation in q t shifts the micro hiring and and layoff relations Fluctuations in aggregate worker flows now arise from shifts in the growth rate distribution and shifts in the micro-level CS relations

10 Exogenously Pro-Cyclical Quit Rates Quit rate drops when aggregate growth rate falls. Inducing rightward shifts in the hiring and layoff relations, including the kink point.

11 Endogenous Quits, 1 Higher Quit Rate at Weaker Employers p Faberman-Nagypál (2008) model n Employers vary in idiosyncratic component of productivity n More productive firms grow faster n Employers engage in costly search, contact workers, and make offers n Bargained wage rises with employer productivity n Because they earn lower wages, workers at less productive employers are more likely to accept outside offers n Thus, quit rate declines with employer growth rates in CS n Rationalizes positive value and a negative slope in the CS hires relation to the left of zero. p See, also, Trapeznikova (2010)

12 Endogenous Quits, 2 Higher Quit Rates in Stronger Labor Market p Barlevy (2002) model with OTJ search n Employed workers quit when better offers arrive n Vacancies are scarcer and workers have fewer outside options in recessions à lower quit rate n Leads to shift and dilation of match quality distribution over business cycle p Shift: negative aggregate shock causes dissolution of bad matches (cleansing effect) p Dilation: lower outside options cause workers in bad matches to remain in those matches (sullying effect) n This model implies that CS quit-growth relation varies with business cycle, shifting up in booms

13 Endogenous Quits, 3 Separation Rates Decline with Job Tenure p Learning about match quality as in Jovanovic (1979, 1985), Moscarini (2005), Pries and Rogerson (2005) and many others n Stochastic match quality n Employer and worker learn about match quality over time n Good matches survive, bad ones don t n Separation rate declines with match tenure n If growing employers have a larger proportion of young matches, then separation rate rises with employer growth rates in the cross section.

14 Relating Micro and Macro Behavior Express aggregate worker flow rates, W t (rate of hires, quits, layoffs or separations), as Group establishments by employment growth rates, g, and calculate the employment-weighted mean rate for each g in period t, w t (g) To recover the aggregate flow rate at t, weight each growth rate bin by its employment mass in period t, f t (g) Obtain w t (g) from JOLTS and f t (g) from BED. (See DFHR, 2010.) Changes over time in aggregate flow rates arise from: 1. Changes in average worker flow rates for a given g, or 2. Shifts in the distribution of establishment-level employment growth 3. Interaction between 1 and 2. W t =! g f t (g)w t (g)

15 Statistical Specifications for CS Relations Fixed Cross-Section Motivated by time-invariant iron link relations in basic multi-worker MP model, but we do not constrain the location of kinks: w t (g) =!(g) + " t D (g) where w t (g) is worker flow rate at establishment with growth rate g Estimate this relation on the pooled sample of establishment-level observations from 2001 to 2010Q2.

16 Percent of Employment C-S Relations in Three Periods Hires Total Separations 2001Q2-2003Q1 2006Q1-2006Q4 2008Q3-2009Q Q2-2003Q1 2006Q1-2006Q4 2008Q3-2009Q2 Percent of Employment Establishment-Level Employment Growth Rate Establishment-Level Employment Growth Rate Layoffs Quits Q2-2003Q1 2006Q1-2006Q4 2008Q3-2009Q2 Percent of Employment Layoffs show stable Iron Layoffs Link relation show stable to employer Iron Link growth relation rates in to the employer CS growth over time Q2-2003Q1 2006Q1-2006Q4 2008Q3-2009Q2 Percent of Employment Quits do not, Quits especially do not, at especially contracting at contracting establishments establishments Establishment-Level Employment Growth Rate Establishment-Level Employment Growth Rate

17 Statistical Specifications, cont d Baseline Allow vertical shifts in CS relation as functions of cycle indicators : w t + ( 4 g) = α ( g) + β1g t + β2gt + β3δgt + β JFt + εt ( g) G t = aggregate employment growth rate (+, -, change) JF t = job-finding rate of unemployed workers Flexible Allow for more complex cyclical behavior Interact cycle indicators with 5 dummy variables for broad growth rate intervals à Allows shape and location of CS relations to vary with cycle. B

18 Worker Flows Implied by Statistical Specifications Wˆ = t! g f t (g)ŵ t (g) Hires Separations Actual Implied by Fixed Cross- Section Relation Implied from Baseline Specification Actual Implied by Fixed Cross- Section Relation Implied from Baseline Specification Layoffs Quits Actual Implied by Fixed Cross- Section Relation Implied from Baseline Specification Actual Implied by Fixed Cross- Section Relation Implied from Baseline Specification

19 160.0 Leftward Shift in Growth Rate Distribution And Interaction with the CS Layoff Relation Percent of Employment Layoff Rate Growth Rate Distribution, Recession Period Growth Rate Distribution, Expansion Period Establishment-Level Employment Growth (Pct. of Employment)

20 How Much Does Fixed CS Model Improve Fit for Aggregate Flows? R-Squared Values in Time-Series Regressions of the Indicated Rate on: Aggregate Variables (4 Cycle Indicators) Hiring Rate Separation Rate Quit Rate Layoff Rate Adding One Variable: Worker Flow Rate Predicted by Fixed CS Model.966 [.000].944 [.000].961 [.011].880 [.000] The entry in brackets reports the p-value of the coefficient on the prediction of the model that imposes a time-invariant cross-sectional relation.

21 Fit of the Establishment-Level Regressions Used to Estimate the CS Worker Flow Relations p Table entries show R-squared values for employmentweighted regressions on the indicated statistical models. n n Fixed Cross-Section corresponds to the regression model used to fit the time-invariant CS relations displayed on the previous slides Augmented Fixed Cross-Section relaxes the model slightly to allow for within-bin differences in the worker flow relations. Dependent variable in descriptive CS regression Fixed Cross- Section Model Specification Augmented Fixed Cross- Section Augmented Baseline Specification Augmented Flexible Specification Hiring Rate Separation Rate Quit Rate Layoff Rate

22 Constructing Synthetic JOLTS Data Baseline statistical model + quarterly data on the cross-sectional distribution of establishment-level growth rates à synthetic data for aggregate worker flows Wˆ = t! g f t (g)ŵ t (g) BED data on f + model-based ŵ from 1990 to 2001 BED data on f + JOLTS-based w from 2001 to 2010.

23 Quit, Layoff, and Job Destruction Rates Synthetic JOLTS Estimates Percent of Employment Job Destruction Layoffs Quits Actual JOLTS Estimates Layoffs move with job destruction. Quits moves opposite to both.

24 Hiring and Job Creation Rates Percent of Employment Synthetic JOLTS Estimates Job Creation (left axis) Hires (right axis) Actual JOLTS Estimates Hires tend to move with job creation but are more volatile. On the secular declines in worker flow rates, see DHJM (2006), Davis (2008) and DFHJM (2010).

25 Overview of Part II 5. Use a simple model of daily hiring dynamics to identify the job-filling rate for vacancies 6. Big CS variation in job-filling rates. Why? n n n Heterogeneity in the efficiency of search and matching Scale economies (or diseconomies) in the hiring technology at the establishment or sectoral level Employers use other instruments, in addition to vacancy numbers, to influence the pace of hiring.

26 Overview of Part II 5. Use a simple model of daily hiring dynamics to identify the job-filling rate for vacancies 6. Big CS variation in job-filling rates. Why? n n n Heterogeneity in the efficiency of search and matching Scale economies (or diseconomies) in the hiring technology at the establishment and/or sectoral Employers use other instruments, in addition to vacancy numbers, to influence the pace of hiring.

27 Overview of Part II 7. Generalized matching function (GMF) defined over unemployment, vacancies, and other recruiting instruments. n n n n n Estimate scale economies in the employer hiring technology Estimate how recruiting intensity varies with hires rate Construct a time-series index of recruiting intensity per vacancy GMF outperforms standard MF in explaining fluctuations in the job-finding rate and the job-filling rate. GMF also yields a more stable Beveridge Curve. GMF accounts for CS behavior of job-filling rates. Standard matching function does not.

28 A Model of Daily Hiring Dynamics Daily laws of mo5on for flow of hires and vacancy stock: h s,t = f t v s!1,t v = [(1! f )(1!! )]v +! s,t t t s!1,t t p Where s indexes days, f t is the daily job- filling rate in month t, δ t is the rate at which unfilled vacancies lapse, and θ t is the daily flow of new vacancies.

29 29

30 Vacancy Flows and Job- Filling Rate Rela5onships to Employer Growth Rates 30

31 Job- Filling Rate and Gross Hires Rate Log Daily Fill Rate Data points correspond to growth rate bins Hires-Weighted Least Squares Slope (s.e.) = (0.006) R-squared = Log Hires Rate

32

33 Quan5fying the Roles of Other Instruments and Scale Economies Let q(v et, x et )! v " et!q(x et ) job- filling rate becomes so that d log( f et ) d log(h et ) = d log( f! t ) d log(h et ) + (!!1 ) d log(v et ) f = f! v!!1!q(x ) et t et et ( ) d log(h et ) + d log!q(x et ) d log(h et ) ( ) = 0 + (!!1)(0.336) + d log!q(x et ) d log(h et ) To preclude a role for employer ac5ons on other margins requires a scale economy parameter value of!!

34 Es5ma5ng Scale Economies in the Establishment- Level Hiring Technology Basic idea: Exploit differences in scale of vacancies and hiring across industry- size cells to es5mate returns to scale in employer hiring technology. Do NOT use 5me varia5on, because it is contaminated by the intensity, x. Control for cell- level growth rate for same reason. Control for differences in matching efficiency across industries and across employer size classes. Instrument using level of employment to deal with poten5al division bias. 34

35 Scale- Economy Regressions!"!!"!!"!!!!!!"!!"!!"!!!!"!!!!" Average Time Effect Mean Job-Filling Rate in Industry i and size class s Sectoral differences in matching efficiency and average recruiting intensity: include industry and size fixed effects and industry-size mean employment growth rates. Mean Number of Vacancies (Stock) In Industry i and size class s Scale-Economy Parameter: Elasticity of Job-Filling Rate with Respect to the Number of Vacancies 35

36 Scale Economies Regressions Dependent Variable: Log(Job- Filling Rate) Explanatory Variable! Log Beginning-of-Month Vacancies (Level) Log Monthly Vacancy Flow (Level) Estimation Method!.OLS IV OLS IV Coefficient (std. error) (.049).001 (.051).065 (.049).001 (.051) R First-stage R Implied! N=70 in all regressions. 5 or 6 size classes per industry (12). 2. All regressions include industry and size class fixed effects and the employment growth rate in the industry-size cell. 3. IV is two-stage LS regression using log(employment Level) as the instrument. N=70 in all regressions. 36

37 Aggregate Implica5ons GMF with CRS at the employer- level implies: H t =! H et = µ e # $ % " v t u t & ' ( )* #! v et!q(x et ) = µ $ % e where q t =!(v et / v t )!q(x et ) and v t " = v t q t. e " v t u t & ' ( )* v t " = µv 1)* t u * t q 1)* t,! log H = "! logu + ( 1# " )! logv + ( 1 # " )! logq Working Hypothesis:! logq! log H =! logq et! log H et =

38 Recrui5ng Intensity Per Vacancy Series Implied by Working Hypothesis 1.2 Effective vacancies equal this index value times the number of measured vacancies January 2001 to December

39 Market Tightness and the Role of Recrui5ng Intensity Per Vacancy mu*(u/v)^0.5 Vacancy Yield (hires per vacancy) The drop in recruiting intensity accounts for one-fourth of the gap that emerges between these two measures from 2007 to Davis (2011) draws on Krueger and Mueller (2011) to develop evidence that a drop in search intensity per unemployed also accounts for a sizable share of the gap that emerges after

40 Testing Performance of Standard vs. GMF In Time-Series Regressions Std. Deviation, Dependent Variable RMSE Using Standard Matching Function Percent Drop in RMSE, Generalized MF Non-Nested Test of Added Predictive Ability p-value, H 0 p-value, H 0 = = Standard Generalized Model Model Specification Job-finding rate (Unemployment Escape Rate) Regressed on Tightness Ratio (v! /u) National Data Job-finding rate (Hires Per Unemployed) Regressed on Tightness Ratio (v! /u) National Data Northeast Midwest South West Unemployment Rate Regressed on Effective Vacancy Rate (v!) National Data Northeast Midwest South West

41 A Summary: Tools and Methods 1. A useful descriptive tool: Relate worker flows and job-filling rates to growth rates in the CS. n Yields empirical objects for assessing, calibrating and developing theory n Highlights the importance of nonlinear aggregation in labor market fluctuations 2. How to combine CS statistical models with administrative data on employer growth rates to construct synthetic data. 3. A simple model + moment-fitting method that identifies job-filling rates from periodic data on the stock of vacancies and the flow of hires

42 A Summary: Tools and Methods 4. A generalized matching function (GMF): n How to estimate the degree of scale economies (or diseconomies) in the employer hiring technology n How to identify the elasticity of recruiting intensity per vacancy with respect to the hires rate n A time-series index for recruiting intensity per vacancy n An aggregate time series for effective vacancies that outperforms the standard measure of vacancies in accounting for fluctuations in job-finding rates and jobfilling rates, and that yields a more stable Beveridge Curve.

43 References (Incomplete) p p p p p p p p p Cooper, Haltiwanger and Willis, 2007, Search Frictions: Matching Aggregate and Establishment-Level Observations, Journal of Monetary Economics, 54, Davis, 2011, Comment on Krueger and Mueller, Brookings Papers on Economic Activity, forthcoming. Davis, Faberman and Haltiwanger, 2010 The Establishment-Level Behavior of Vacancies and Hiring, NBER Working Paper Revision coming soon. Davis, Faberman and Haltiwanger, 2011, Labor Market Flows in the Cross Section and Over Time, NBER w.p. no Journal of Monetary Economics, forthcoming. Davis, Faberman, Haltiwanger and Rucker, 2010, Adjusted Estimates of Worker Flows and Job Openings in JOLTS. In Abraham, Harper and Spletzer, eds., Labor in the New Economy, University of Chicago Press, Elsby and Michaels, 2011, Marginal Jobs, Heterogeneous Firms and Unemployment Flows, working paper. Krueger and Mueller, 2011, Job Search, Emotional Well-Being and Job Finding in a Period of Mass Unemployment, Brookings Papers on Economic Activity, forthcoming. Mortensen and Pissarides, 1994, Job Creation and Job Destruction in the Theory of Unemployment, Review of Economic Studies, 61, no. 3, Trapeznikov, Ija, 2010, Employment Adjustment and Labor Utilization, working paper, Northwestern University.

44 ADDITIONAL SLIDES TO BE DISCUSSED AS TIME PERMITS

45 U.S. Employment Growth Rate Distribution, Selected Periods, BED Data Fraction of Employment at q2-2003q q3-2009q2 Establishments with Contractions > 10%, including Closings Establishments with Contractions >= 10% Establishments with No Net Change in Employment Establishments with Expansions < 10% Establishments with Expansions >=10%, including Openings

46 Layoff Rates Compared to Other Job Loss Data Job Destruction (left axis) Layoffs (Left Axis) Unemployment Inflows, CPS (Left Axis) Initial UI Claims (Right Axis) Percent of Employment

47 Job- Filling Rate and Gross Hires Rate T= Turnover Quintile I=Industry S=Size Class 47

48 Is It Just Lucky Employers Growing Faster? Stochas5c nature of job filling induces a posi5ve rela5onship between realized employment growth and job- filling rates at the establishment level. Lucky employers fill jobs faster and, as a result, grow faster. To quan5fy this effect, we simulate hires and employment growth at the establishment level for fiwed values of f, θ, δ, and the distribu5on of vacancies, allowing parameters and vacancy distribu5ons to vary freely by employer size class. Result: Luck effect is much too small to explain the observed C- S rela5onship between job- filling rate and growth rate: Luck alone à job- filling rate rises by 2 percentage points in moving from 0% to 10% monthly growth rate. It rises by another 1 point in moving from 10 to 30%. 48

49 Simulated and Empirical Job-Filling Rates Compared!"'!#!"&$#!"&!#!"%$#!"##$%&'($ )*+,-,./0#123(4,00,56#7/89# D-2+2-E25#82#)A8/30,AF*958#)*+02G*958#!"%!#!"!$#!"!!# ('!"!# (&$"!# (&!"!# (%$"!# (%!"!# ($"!#!"!# $"!# %!"!# %$"!# &!"!# &$"!# '!"!# )*+',#-$./0#*-/(+'$12*3',$%&'($45(26(+'7$

50 Closely Related Work in Progress p Apply the same statistical approach to the analysis of vacancies: n Assess theoretical models n Construct synthetic JOLTS-like vacancy measures back to 1990 n Construct highly disaggregated vacancy measures by region, industry, employer size, etc. (with the intention to overcome smallsample problems in disaggregated vacancy measures calculated directly from JOLTS).

51 Textbook Equilibrium Search Model No role for recrui5ng intensity per vacancy Pissarides (2000, chapter 5) extends standard model to incorporate variable recrui5ng intensity per vacancy Costs per vacancy are increasing and convex in intensity His hiring technology and matching func5on are consistent with our generalized matching func5on (micro CRS case) Op&mal recrui&ng intensity is insensi&ve to aggregate condi&ons and same for all employers in the cross- sec&on. Why? Employers use vacancies to vary hires, and choose intensity to minimize cost per vacancy. Rejected by our CS evidence, specifically posi5ve rela5onship of job- filling rates to employer growth and hires rate. Cannot explain role of other instruments for aggregate hires. 51

52 Addi5onal Theore5cal Implica5ons A major role for recrui5ng intensity per vacancy is not fatal to standard equilibrium search models with random matching, but it calls for re- evalua5on of widely used building blocks in the standard model Dropping the standard free- entry condi5on for new jobs (and dispensing with the convenient result that equilibrium vacancy value is 0) leads to a meaningful role for recrui5ng intensity per vacancy. See Davis (2001), Quality Distribu5on of Jobs The CS evidence on slides is hard to square with the basic mechanism stressed by mismatch models. Directed search models are readily compa5ble with the CS evidence, because these models come built- in with an extra recrui5ng margin, typically in the form of posted offer wages. See Kass and Kircher (2010). 52

53 Are All Hires Mediated through Vacancies? A Test 53

54 Model Specifica5on Test Results Percent of Hires in t by Establishments with No Vacancies at end of t Percent Implied by Model for Alterna5ve Sectoral Breakdowns Size Class (6) by Worker Turnover Rate (6) 36 cells 27.0 Industry (12) by Size Class (2) by Worker Turnover (6) Rate 144 cells 26.7 Industry (2) by Size Class (6) by Worker Turnover Rate (15) 180 cells /41.6 = 66% à Our model of daily hiring accounts for about 2/3 of hires at establishments with no vacancies at start of month. So a big share of hires are not mediated through vacancies 54

55 55